Hi Alun and Kay,


PDB_REDO does some more extended twinning testing with among others Phaser, so 
it would be worthwhile to have a look at that. On the server it refuses to do 
twinned refinement if there is no clear indication of twinning even if the user 
assumed twinning before.

The fact the R-free does nothing, but R goes down using the twin target 
suggests me that there is more going on in the refinement. Perhaps you should 
look at optimising other refinement parameters*



Cheers,

Robbie



*e.g. with The pdb-redo webserver ?



Sent from my Windows 10 phone



Van: Kay Diederichs<mailto:kay.diederi...@uni-konstanz.de>
Verzonden: woensdag 1 maart 2017 23:38
Aan: CCP4BB@JISCMAIL.AC.UK<mailto:CCP4BB@JISCMAIL.AC.UK>
Onderwerp: Re: [ccp4bb] Twinning and R-Free



Hi Alun,

that's difficult to understand for me. It is my understanding that for an 
imperfect model, refmac will indicate a twin fraction >0 even if in reality it 
is 0. But alpha=0.372 sounds too high for that.
My questions would be -
a) is it a single dataset, or did you merge ? If the latter, it might just be a 
problem of alternative indexing.
b) PDB_REDO uses refmac AFAIK so the result should not be different from when 
_you_ run refmac?
c) the spacegroup is I4 but twinning makes it appear close to I422?
d) what exactly are the aimless statistics you refer to? what exactly does 
pointless report?
e) could it be a data processing problem? Crummy data often look twinned just 
because errors are interpreted as twinning.

best,
Kay



On Wed, 1 Mar 2017 22:03:07 +0000, Alun R Coker <alun.co...@ucl.ac.uk> wrote:

>Hi Everyone,
>
>I have a (2.5 - 2.3 Angstroms) data set the process as I4 or I222.
>Aimless indicates twinning in both space groups and all the lower
>symmetry space groups consistent with the reduced cell. Molecular
>replacement works in I4 but not I222 etc. I can see new density for a
>ligand we are interested in some of the 6 subunits so all seems
>promising. I've refined in refmac without the twinning turned on in
>order to avoid model bias and end up with an R-factor/R-free of
>0.268/0.305. However, if I repeat the last round of refinement with
>twinning turned on the R-factor drops but not the R-free (R/R-free
>0.237/0.305). Remac reckons the twin fraction is 0.628/0.372.
>
>My question is does the drop in R-factor but not R-free suggest I am
>over-fitting the model by refining twinning?
>
>Running the data and model through pdb redo suggest the data isn't twinned.
>
>Thanks,
>
>Alun
>
>--
>Dr Alun R. Coker
>Senior Lecturer
>Wolfson Institute for Biomedical Research
>University College London
>The Cruciform Building
>London
>WC1E 6BT
>
>Tel: 020 7679 6703 Ext 46703
>Web: www.ucl.ac.uk/pxmed<http://www.ucl.ac.uk/pxmed>

Reply via email to